28

How to construct/describe the coproduct of two - not necessarily commutative - rings $R$ and $S$?

This in category $\mathbf{Ring}$ having as objects rings with a unit and as arrows unitary ringhomomorphisms.

I thought of firstly constructing monoid $M$ as coproduct of the underlying monoids $U(R)$ and $U(S)$ where $U:\mathbf{Ring}\rightarrow\mathbf{Mon}$ denotes the forgetful functor, and then secondly taking the ring $\mathbb{Z}\left[M\right]$ free over monoid $M$, but still have my doubts. If the rings have finite coprime characteristics then the coproduct should be the trivial ring, so something is wrong.

Can you give me a description of the coproduct (including its injections)? Thank you in advance.

Smiley1000
  • 4,219
drhab
  • 153,781
  • just for comparison, http://math.stackexchange.com/questions/143098/coproduct-in-the-category-of-noncommutative-associative-algebras – Noix07 Aug 05 '14 at 14:43
  • Related questions: https://math.stackexchange.com/q/197819/491450 , https://math.stackexchange.com/q/143098/491450 , https://math.stackexchange.com/q/1972781/491450 , https://math.stackexchange.com/q/3865235/491450 , https://math.stackexchange.com/q/2016341/491450 – Smiley1000 Dec 18 '24 at 19:15

2 Answers2

35

The coproduct of rings (with $1$) looks somewhat similar to the coproduct of groups or monoids (aka free products), but that we use $\otimes_\mathbb{Z}$ instead of $\times$. In fact, both are special cases of a general construction which works for monoid objects in arbitrary cocomplete monoidal categories.

The coproduct $R \sqcup S$ of two rings $R,S$ is constructed as follows: Let us denote the underlying abelian groups also as $R,S$. Then consider the direct sum of tensor products of abelian groups

$\mathbb{Z} \oplus R \oplus S \oplus \bigl(R \otimes S\bigr) \oplus \bigl(S \otimes R\bigr) \oplus (R \otimes S \otimes R) \oplus \bigl(S \otimes R \otimes S\bigr) \oplus \cdots$

Now we quotient out the relations $$x_1 \otimes \cdots \otimes x_n \equiv x_1 \otimes \cdots \otimes x_n \otimes 1 \equiv 1 \otimes x_1 \otimes \cdots \otimes x_n,$$ $$\cdots \otimes x_i \otimes 1 \otimes x_{i+1} \otimes \cdots \equiv \cdots \otimes x_i x_{i+1} \otimes \cdots.$$ The quotient has a multiplication characterized by $$(x_1 \otimes \cdots \otimes x_n ) \cdot (y_1 \otimes \cdots \otimes y_m) := $$ $$\left\{\begin{array}{ll} x_1 \otimes \cdots \otimes x_n \otimes y_1 \otimes \cdots \otimes y_m & x_n \in R, y_1 \in S \text{ or } x_n \in S, y_1 \in R \\ x_1 \otimes \cdots \otimes x_n y_1 \otimes \cdots \otimes y_m & x_n,y_1 \in R \text{ or } x_n,y_1 \in S\end{array}\right.$$ We obtain a ring $R \sqcup S$ with homomorphisms $R \rightarrow R \sqcup S \leftarrow S$. It is the coproduct: Given homomorphisms $f : R \to T$ and $g : S \to T$, we define a homomorphism $h : R \sqcup S \to T$ by mapping, for example $x_1 \otimes x_2 \otimes x_3 \in R \otimes S \otimes R$ to $f(x_1) \cdot g(x_2) \cdot f(x_3)$. This is clearly a homomorphism of abelian groups on the infinite direct sum, but it respects the relations and therefore extends to a homomorphism of abelian groups on the quotient. It is the unique ring homomorphism satisfying $h|_R = f$ and $h|_S =g$.

The construction of $R \sqcup S$ is somewhat complicated, but notice that its elements are just formal sums of products taken from $R$ or $S$, for example $r_1 + s_1 \cdot r_2 - r_3 \cdot s_2 \cdot r_4$. This is how one comes up with the construction.

If $k$ is a commutative ring, then coproducts of $k$-algebras (with $1$) can be described exactly in the same way, we just take $\otimes_k$ instead of $\otimes_{\mathbb{Z}}$. The coproduct of rngs (rings without $1$) is also similar, even a bit easier, see here.

As with all algebraic structures, coproducts can also be described using generators and relations: If $R \cong \mathbb{Z}\langle X \rangle / I$ and $S \cong \mathbb{Z} \langle Y \rangle / J$ are quotients of free rings (for sets of variables $X,Y$ and ideals $I,J$), then $$R \sqcup S = \mathbb{Z} \langle X \sqcup Y \rangle / (I+J).$$ This description is very useful when presentations of $R$ and $S$ are known. For example, we get $\mathbb{Z} \langle x,y \rangle / (x^2-y^3) \sqcup \mathbb{Z} \langle u,v,w \rangle / (u^2-vw) = \mathbb{Z} \langle x,y,u,v,w \rangle / ( x^2-y^3, u^2-vw)$. Basically, this approach uses the forgetful functor $\mathbf{Ring} \to \mathbf{Set}$ to construct coproducts of rings with the help of coproducts of sets.

Your idea, using the forgetful functor $U : \mathbf{Ring} \to \mathbf{Mon}$ (forgets addition) and its left adjoint $\mathbb{Z}[-]$, the monoid algebra (over $\mathbb{Z}$), can also be made to work: We can construct the coproduct of two rings $R,S$ as $\mathbb{Z}[U(R) \sqcup U(S)]/I$, where the ideal $I$ is generated by the relations $[r+r'] \equiv [r] + [r']$ for $r,r' \in U(R)$ (where $[r]$ denotes the image of $r \in U(R)$ in the monoid ring) and $[s+s'] \equiv [s] + [s']$ for $s,s' \in U(S)$. These relations exactly guarantee that the cospan of monoid homomorphisms $$U(R) \to U(\mathbb{Z}[U(R) \sqcup U(S)]) \leftarrow U(S)$$ lifts to a cospan of ring homomorphisms $$R \to \mathbb{Z}[U(R) \sqcup U(S)]/I \leftarrow S.$$

But we may also use the forgetful functor $V : \mathbf{Ring} \to \mathbf{Ab}$ (forgets multiplication) and its left adjoint $T : \mathbf{Ab} \to \mathbf{Ring}$, the tensor algebra (over $\mathbb{Z}$). We construct the coproduct of two rings $R,S$ as $T(V(R) \oplus V(S)) / J$, where $J$ is the ideal generated by the relations $[1_R] \equiv 1$, $[rr'] \equiv [r] [r']$, $[1_S] \equiv 1$, $[ss'] \equiv [s] [s']$. These relations exactly guarantee that the cospan of abelian group homomorphisms $$V(R) \to V(T(V(R) \oplus V(S))) \leftarrow V(S)$$ lifts to a cospan of ring homomorphisms $$R \to T(V(R) \oplus V(S)) / J \leftarrow S.$$ More generally, if $\tau' \to \tau$ is any morphism of algebraic theories, there is a forgetful functor $U : \mathbf{Alg}(\tau) \to \mathbf{Alg}(\tau')$ with a left adjoint $F$, and colimits in $\mathbf{Alg}(\tau)$ can then be constructed using colimits in $\mathbf{Alg}(\tau')$ by generalizing the construction above, namely $\mathrm{colim}_i A_i$ for $\tau$-algebras $A_i$ is a suitable quotient algebra of $F(\mathrm{colim}_i U(A_i))$.

  • Is what you call 'algebraic structure' the same thing as 'algebraic system (of a given type)' mentioned in CWM on page 120? By the way, you are of great help to me on this site. Not only in this question. Tibi gratias ago! – drhab Jan 03 '14 at 14:44
  • Yes, exactly. CWM keeps this quite short (although very concise), you can find more about algebraic structures in texts about universal algebra. – Martin Brandenburg Jan 03 '14 at 14:46
  • 1
    I don't think $R$ and $S$ embed into your coproduct in the sense of injective maps. Try $S = 0$ and $R \neq 0$. – darij grinberg Jan 03 '14 at 15:25
  • @Darij: You are right, thank you. – Martin Brandenburg Jan 03 '14 at 15:27
  • @MartinBrandenburg: I'm guessing the same construction works for $R$-algebras: $$A\ast B= \frac{R\oplus A\oplus B\oplus (A!\otimes_R!A)\oplus (A!\otimes_R!B)\oplus (B!\otimes_R!A)\oplus (B!\otimes_R!B)\oplus \ldots}{ \langle\ldots!\otimes! a!\otimes! a'!\otimes!\ldots!-!\ldots!\otimes! aa'!\otimes\ldots,: \ldots!\otimes! b!\otimes! b'!\otimes!\ldots!-!\ldots!\otimes! bb'!\otimes!\ldots,: 1_R!-!1_A,: 1_R!-!1_B\rangle}$$ and $s\cdot t:=!s!\otimes!t$ for $s,t!\in!A!\ast!B$, right? If $A$ and $B$ are graded, how is $A!\ast!B$ graded? – Leo Mar 13 '15 at 19:41
  • I'm assuming you meant alternating summands $A!\otimes!B!\otimes!A!\otimes!B!\otimes!A!\otimes!B!\otimes!\ldots$ and $B!\otimes!A!\otimes!B!\otimes!A!\otimes!B!\otimes!A!\otimes!\ldots$. But aren't you missing relations $1_R-1_A$ and $1_R-1_B$? Or is the copy of $R$ perhapy unnecessary, since we have $1_A=1_A\otimes1_B=1_B$? – Leo Mar 13 '15 at 19:54
  • @Martin, what if $R$ and $S$ are assumed to have nonzero intersection $T$? Is the definition of coproduct detailed enough so that it recognizes this fact and reduces the computation when an element of $R$ meets an element of the copy of $T$ which lies inside $S$, or is it more crude and the two copies of $T$ (one inside $R$, one inside $S$) are in fact two cartesianly separated objects? – Jose Brox Jun 25 '15 at 12:40
  • 1
    It doesn't make sense to talk about $R \cap S$. And yes, $R$ and $S$ are completely isolated. – Martin Brandenburg Jun 25 '15 at 12:59
  • @MartinBrandenburg (Sorry, didn't see your answer!!). From a ring-theoretic perspective I think it makes sense to talk about $R\cap S$ when we have an ambient ring as universe (e.g $R$ upper triangular matrices, $S$ lower triangular matrices, $D=R\cap S$ diagonal matrices, all inside a matrix ring). This still bugs me: the ring $R\coprod S/I$, where $I$ is the ideal that mods out $R\cap S$ (i.e., $x\oplus 0=0\oplus x$ when $x\in R\cap S$ and derived identities), seems more natural to me, fusing the two rings where the two overlap. Do you know if this construction has any name? Or has problems? – Jose Brox Nov 07 '17 at 17:19
  • I find this description very useful. However, and moving to the case of algebras, I do not see how this could be the same of the tensor product of commutative algebras. It looks more like the tensor algebra rather than the tensor product of algebras. – gibarian Oct 15 '20 at 07:44
  • @gibarian [I do not see how this could be the same of the tensor product of commutative algebras.]---> Not at the level of the construction but at the level of universal solution, changing the ambient category. – Duchamp Gérard H. E. Feb 19 '21 at 07:13
  • @darijgrinberg Embedding (resp. universal solution) works for augmented algebras (resp. within the category of augmented algebras). – Duchamp Gérard H. E. Feb 19 '21 at 07:23
  • Isn't it simpler to say that if $A,B$ are two $R$-algebras (they could be $\mathbb Z$-algebras so that covers the case of rings), you can define their coproduct as a quotient of $$ A \sqcup B = \bigoplus_{n \ge 0} (A \oplus B)^{\otimes n} / rels $$ where $(A \oplus B)^{\otimes 0} = R$ and all the tensor products are taken as $R$-modules? If we have a third $R$-algebra $C$ with $R$-algebra morphisms $A,B \to C$, we get a map of $R$-modules $A \oplus B \to C$, extend it to $(A \oplus B)^{\otimes n} \to C$ by taking tensor powers of the same map, and then their direct sum $A \sqcup B \to C$. – Patrick Da Silva Aug 28 '23 at 18:50
  • The ideal $rels$ (ran out of characters) would be basically defined by the relations which are trivially holding when we map all those elements into $C$, such as the ones already mentioned, so you get generators of the form $$ a_1 \otimes a_2 - a_1a_2, \quad a_1, a_2 \in A \cdots \text{ same with B } $$ or $$ a \otimes 1 - a, \quad b \otimes 1 - b, \quad a \in A, b \in B $$ – Patrick Da Silva Aug 28 '23 at 18:57
  • @JoseBrox from the past, the concept you were looking for is the pushout. – Jose Brox Feb 02 '24 at 22:17
  • @PatrickDaSilva Yes, sure. You take the tensor algebra (=free algebra) over the coproduct of abelian groups, then mod out the necessary relations. This works for every forgetful functor as well. My construction is a bit more explicit, though. – Martin Brandenburg Aug 11 '24 at 12:08
  • @PatrickDaSilva I have revised and improved the answer. Your description is now also included (but formulated more abstractly). – Martin Brandenburg Sep 25 '24 at 18:16
1

Here is a way to construct the coproduct of a family of rings $R_i$ in the category $\mathsf{Ring}$:

Let $S = \mathbb{Z} \langle x_{i, r} : i \in I, r \in R_i \rangle$ be the free noncommutative ring on the indeterminates $x_{i, r}$ for $i \in I$ and $r \in R_i$. Now let $J \subseteq S$ be the ideal generated by expressions of the following form:

  • $x_{i, r_1} + x_{i, r_2} - x_{i, r_1 + r_2}$ for $i \in I$ and $r_1, r_2 \in R_i$
  • $x_{i, r_1} x_{i, r_2} - x_{i, r_1 r_2}$ for $i \in I$ and $r_1, r_2 \in R_i$
  • $x_{i, 1} - 1$ for $i \in I$

Now take $\newcommand\quotient[2]{{^{\Large #1}}/{_{ \Large #2}}} T = \quotient{S}{J}$. Then $T \cong \coprod_{i \in I} R_i$ is the coproduct over all the $R_i$ in $\mathsf{Ring}$. The inclusions $R_i \to T$ are the obvious ones, mapping $r$ to $x_{i, r}$, and the fact that we took the quotient by $J$ ensures that these are actually ring homomorphisms.


Here are some related questions:

Smiley1000
  • 4,219
  • 1
    This is already contained in my answer. Read the section "As with all algebraic structures, coproducts can also be described using generators and ..." – Martin Brandenburg Dec 18 '24 at 17:54